Susindra Reddy
I work on LLM post-training and reinforcement learning for agents.
I built an asynchronous RL pipeline with Prime-RL and Modal to train Qwen3.5-9B using three H100s, covering inference benchmarks, an 89% reduction in minimum sandbox costs, and an investigation into why training hurt performance.
I also fine-tuned Qwen3.8-27B to compose music using verified responses from GPT-6.1 Sol, with QLoRA response distillation. Read the write-up or explore the recorded training run, with rewards, logs, and music samples.
Agentmeets is a meeting platform where people and an AI teammate talk, research the web, and share notes in one room.
I also built Nexus Medical, a platform for medical exam prep and practice consultations with AI voice agents.
Find the Beat generates musical accompaniment and sheet music from live audio signals.
Job Agent uses LangGraph to help find jobs and manage applications.
BMR Buzz supports AI-assisted writing and publishing.
I’m a master’s student in Data Science and AI at the University of Houston, where I also work as an Instructional Assistant. Previously, I worked at Pravaah Consulting as an AI/ML intern and later as a developer.
At Pravaah, I received Best Intern (2023) and Innovator of the Year (2024). I also earned 3rd place at Anveshana National Hackathon (2022).